@llm-dev-ops/agentics-cli 1.4.53 → 1.4.55

package/dist/bundled-agents/copilot-agents/functions/dist/services/agents/intent-classifier/intent-classifier-agent.d.ts

@@ -0,0 +1,123 @@
+/**
+ * Intent Classifier Agent
+ *
+ * Purpose: Classify user or system intent to guide downstream reasoning
+ * Classification: INTENT_ANALYSIS
+ * decision_type: intent_classification
+ *
+ * Scope:
+ * - Classify intent type
+ * - Detect multi-intent states
+ * - Assign confidence scores
+ *
+ * CONSTITUTION COMPLIANCE:
+ * ✓ Stateless at runtime
+ * ✓ Emits exactly ONE DecisionEvent per invocation
+ * ✓ Persists ONLY via ruvector-service
+ * ✓ NEVER connects directly to databases
+ * ✓ NEVER executes SQL
+ * ✓ NEVER modifies runtime behavior
+ * ✓ NEVER orchestrates other agents
+ * ✓ NEVER enforces policy
+ * ✓ NEVER intercepts execution paths
+ *
+ * Must Never:
+ * - Trigger workflows
+ * - Route execution
+ * - Enforce policy
+ */
+import { BaseAgent, AgentMetadata, AgentResult } from '../contracts';
+import { IntentClassifierInput, IntentClassifierOutput } from '../contracts/intent-classifier-schemas';
+import { RuvectorPersistence } from '../planner/ruvector-persistence';
+import { Telemetry } from '../planner/telemetry';
+/**
+ * Intent Classifier Agent Implementation
+ *
+ * This agent analyzes text and classifies user/system intent.
+ * It is purely analytical - it NEVER executes, routes, or enforces anything.
+ */
+export declare class IntentClassifierAgent implements BaseAgent<IntentClassifierInput, IntentClassifierOutput> {
+    readonly metadata: AgentMetadata;
+    private readonly persistence;
+    private readonly telemetry;
+    constructor(persistence: RuvectorPersistence, telemetry: Telemetry);
+    /**
+     * Validate input against IntentClassifierInputSchema
+     */
+    validateInput(input: unknown): IntentClassifierInput;
+    /**
+     * Invoke the intent classifier agent
+     *
+     * DETERMINISTIC: Same input always produces same output structure
+     * STATELESS: No internal state modified
+     * NON-BLOCKING: Fully async
+     */
+    invoke(input: IntentClassifierInput, executionRef: string): Promise<AgentResult>;
+    /**
+     * Classify intent from input text
+     *
+     * This is the core classification logic - purely analytical.
+     * NEVER triggers workflows, routes execution, or enforces policy.
+     */
+    private classifyIntent;
+    /**
+     * Normalize text for analysis
+     */
+    private normalizeText;
+    /**
+     * Detect all potential intents from normalized text
+     */
+    private detectIntents;
+    /**
+     * Extract target from text based on intent
+     */
+    private extractTarget;
+    /**
+     * Extract action from text based on intent
+     */
+    private extractAction;
+    /**
+     * Extract scope from input context
+     */
+    private extractScope;
+    /**
+     * Apply hints to filter intents
+     */
+    private applyHints;
+    /**
+     * Create unknown intent fallback
+     */
+    private createUnknownIntent;
+    /**
+     * Analyze multi-intent state
+     */
+    private analyzeMultiIntentState;
+    /**
+     * Detect relationship between intents
+     */
+    private detectIntentRelationship;
+    /**
+     * Detect ambiguity in classification
+     */
+    private detectAmbiguity;
+    /**
+     * Detect language of input
+     */
+    private detectLanguage;
+    /**
+     * Generate analysis notes
+     */
+    private generateNotes;
+    /**
+     * Calculate overall confidence
+     */
+    private calculateOverallConfidence;
+    /**
+     * Get constraints applied during classification
+     */
+    private getAppliedConstraints;
+    /**
+     * Classify error for proper error code
+     */
+    private classifyError;
+}